Policy Search-Based Distributed Model Predictive Contouring Control of Automated Guided Vehicles in Dynamic Environment

Considering the trajectory following control problem for automated guided vehicles (AGVs) in a dynamic environment with multiple pedestrians, we propose a novel policy search-based distributed model predictive contouring control (PS-DMPCC) scheme to improve the adaptability of controller by searching for the optimal control hyper-parameters in real-time according to observed obstacle states. In traditional cooperative control strategies, robot formations and cost weights are typically designed in advance, leading to challenges in achieving a balance between obstacle avoidance and formation maintenance. Therefore, this paper treats the formation and cost weights as hyper-parameters of MPCC and employs an expectation maximization-based policy search strategy to automatically select optimal hyper-parameters, thereby maintaining expected cooperative control performance in the dynamic environment. Finally, simulation experiments of multiple AGVs in a dense pedestrian environment verify the efficacy of the proposed PS-DMPCC scheme.